New Framework for Self-Supervised Learning in PPG-Based Health Monitoring
Researchers have proposed TS2TC, a generative self-supervised representation learning framework designed for non-invasive estimation of physiological parameters using photoplethysmography (PPG) data. The framework addresses challenges in aligning labels with large-scale PPG data for deep learning, which is resource-intensive. TS2TC leverages temporal, spectrogram, and temporal-spectrogram mixed domains to learn robust shared representations from unlabeled data. A key component is the Cross-Temporal Fusion Generative Anchor (CTFGA) pretext task, which models temporal dependencies and reconstructs independent segments at a coarse level for global feature extraction and local contextual representation. The framework aims to enable universal and noninvasive physiological parameter estimation. The research is detailed in arXiv preprint 2604.22780.
Key facts
- TS2TC is a generative self-supervised learning framework for PPG data.
- It uses temporal, spectrogram, and temporal-spectrogram mixed domains.
- The CTFGA pretext task models temporal dependencies and reconstructs segments.
- The framework aims for universal noninvasive physiological parameter estimation.
- The research is published on arXiv with ID 2604.22780.
- Self-supervised learning helps handle limited annotated data.
- PPG data is used for photoplethysmography-based health monitoring.
- The framework addresses challenges in aligning labels with large-scale data.
Entities
Institutions
- arXiv